Merge pull request #168 from matchms/dev_pytorch

Dev pytorch

.github/workflows/CI_build.yml

@@ -44,7 +44,7 @@ jobs:
       fail-fast: false
       matrix:
         os: ['ubuntu-latest', 'macos-latest', 'windows-latest']
-        python-version: ['3.8', '3.9', '3.10']
+        python-version: ['3.9', '3.10', '3.11']
         exclude:
           # already tested in first_check job
           - python-version: 3.9
@@ -69,38 +69,3 @@ jobs:
       - name: Run tests
         run: |
           pytest
-
-  tensorflow_check:
-    name: Tensorflow version check / python-3.8 / ubuntu-latest
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v2
-      - name: Set up Python
-        uses: actions/setup-python@v1
-        with:
-          python-version: 3.8
-      - name: Python info
-        run: |
-          which python
-          python --version
-      - name: Install Tensorflow version 2.6
-        run: |
-          python -m pip install --upgrade pip
-          pip install "tensorflow>=2.6,<2.7"
-      - name: Install other dependencies
-        run: |
-          pip install -e .[dev,train]
-      - name: Show pip list
-        run: |
-          pip list
-      - name: Run test with tensorflow version 2.6
-        run: pytest
-      - name: Install Tensorflow version 2.8
-        run: |
-          pip install --upgrade "numpy<1.24.0"
-          pip install --upgrade "tensorflow>=2.8,<2.9"
-      - name: Show pip list
-        run: |
-          pip list
-      - name: Run test with tensorflow version 2.8
-        run: pytest

.gitignore

@@ -18,6 +18,7 @@ xunit-result.xml
 
 docs/_build
 docs/apidocs
+prototyping/
 
 # ide
 .idea

CHANGELOG.md

@@ -5,6 +5,26 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.0.0] - date...
+Large scale expansion, revision, and restructuring of MS2Deepscore.
+
+### Added
+- Models are now build using PyTorch.
+- Models have build-in GPU support (using pytorch).
+- new `EmbeddingEvaluatorModel` (Inception Time CNN)
+- new `LinearModel` for absolute error estimates
+- new `MS2DeepScoreEvaluated` matchms-style score --> gives "score" and "predicted_absolute_error"
+- Additional smart binning layer that can handle input of much higher peak resolution (not used as a default!)
+- New validation concept --> all-vs-all scores for the validation spectra are computed, but loss is then computed per score bin. This gives better and more significant statistics of the model performance
+- New loss functions "Risk Aware MAE" and "Risk Aware MSE" which function similar to MAE or MSE but try to counteract the tendency of a model to predict towards 0.5.
+- Losses can now be weighted with a weighting_factor.
+
+
+### Changed
+- No longer supports Tensorflow/Keras
+- The concept of Spectrum binning has changed and is now implemented differently (i.e. no more "missing peaks" as before)
+- Monte-Carlo Dropout does not return a score (mean or median) together with percentile-based upper and lower bound (instead of STD or IQR before).
+
 ## [Unreleased]
 
 ## [1.0.0] - 2024-03-12

README.md

@@ -116,11 +116,13 @@ In that scenario, `scores["score"]` contains the similarity scores (median of th
 Training your own model is only recommended if you have some familiarity with machine learning. 
 To train your own model you can run the code below.
 Please first ensure cleaning your spectra. We recommend using the cleaning pipeline in [matchms](https://github.com/matchms/matchms).
+
 ```python
-from ms2deepscore.train_new_model.SettingMS2Deepscore import \
+from ms2deepscore.SettingsMS2Deepscore import
     SettingsMS2Deepscore
-from ms2deepscore.wrapper_functions.training_wrapper_functions import \
+from ms2deepscore.wrapper_functions.training_wrapper_functions import
     train_ms2deepscore_wrapper
+
 settings = SettingsMS2Deepscore({"epochs": 300,
                                  "base_dims": (1000, 1000, 1000),
                                  "embedding_dim": 500,
@@ -129,9 +131,7 @@ settings = SettingsMS2Deepscore({"epochs": 300,
                                  "learning_rate": 0.00025,
                                  "patience": 30,
                                  })
-train_ms2deepscore_wrapper(spectra_file_path=, 
-                           settings=settings, 
-                           validation_split_fraction=20)
+train_ms2deepscore_wrapper(spectra_file_path=, model_settings=, validation_split_fraction=20)
 ```
 ## Contributing
 We welcome contributions to the development of ms2deepscore! Have a look at the [contribution guidelines](https://github.com/matchms/ms2deepscore/blob/main/CONTRIBUTING.md).

ms2deepscore/MS2DeepScore.py

@@ -2,8 +2,8 @@
 import numpy as np
 from matchms import Spectrum
 from matchms.similarity.BaseSimilarity import BaseSimilarity
-from tqdm import tqdm
-from .typing import BinnedSpectrumType
+from ms2deepscore.models.SiameseSpectralModel import (SiameseSpectralModel,
+                                                      compute_embedding_array)
 from .vector_operations import cosine_similarity, cosine_similarity_matrix
 
 
@@ -29,7 +29,7 @@ class MS2DeepScore(BaseSimilarity):
         queries = load_from_json("xyz.json")
 
         # Load pretrained model
-        model = load_model("model_file_123.hdf5")
+        model = load_model("model_file_123.pt")
 
         similarity_measure = MS2DeepScore(model)
         # Calculate scores and get matchms.Scores object
@@ -38,43 +38,25 @@ class MS2DeepScore(BaseSimilarity):
 
     """
 
-    def __init__(self, model, progress_bar: bool = True):
+    def __init__(self, model: SiameseSpectralModel, progress_bar: bool = True):
         """
 
         Parameters
         ----------
         model:
             Expected input is a SiameseModel that has been trained on
-            the desired set of spectra. The model contains the keras deep neural
-            network (model.model) as well as the used spectrum binner (model.spectrum_binner).
+            the desired set of spectra.
         progress_bar:
             Set to True to monitor the embedding creating with a progress bar.
             Default is False.
         """
         self.model = model
-        self.multi_inputs = (model.nr_of_additional_inputs > 0)
-        if self.multi_inputs:
-            self.input_vector_dim = [self.model.base.input_shape[0][1], self.model.base.input_shape[1][1]]
-        else:
-            self.input_vector_dim = self.model.base.input_shape[1]
-        self.output_vector_dim = self.model.base.output_shape[1]
+        self.model.eval()
+        self.output_vector_dim = self.model.model_settings.embedding_dim
         self.progress_bar = progress_bar
 
-    def _create_input_vector(self, binned_spectrum: BinnedSpectrumType):
-        """Creates input vector for model.base based on binned peaks and intensities"""
-        if self.multi_inputs:
-            X = [np.zeros((1, i[1])) for i in self.model.base.input_shape]
-            idx = np.array([int(x) for x in binned_spectrum.binned_peaks.keys()])
-            values = np.array(list(binned_spectrum.binned_peaks.values()))
-
-            X[0][0, idx] = values
-            X[1] = np.array([[float(value) for key, value in binned_spectrum.metadata.items() if (key != "inchikey")]])
-        else:
-            X = np.zeros((1, self.input_vector_dim))
-            idx = np.array([int(x) for x in binned_spectrum.binned_peaks.keys()])
-            values = np.array(list(binned_spectrum.binned_peaks.values()))
-            X[0, idx] = values
-        return X
+    def get_embedding_array(self, spectrums):
+        return compute_embedding_array(self.model, spectrums)
 
     def pair(self, reference: Spectrum, query: Spectrum) -> float:
         """Calculate the MS2DeepScore similaritiy between a reference and a query spectrum.
@@ -91,12 +73,9 @@ def pair(self, reference: Spectrum, query: Spectrum) -> float:
         ms2ds_similarity
             MS2DeepScore similarity score.
         """
-        binned_reference = self.model.spectrum_binner.transform([reference])[0]
-        binned_query = self.model.spectrum_binner.transform([query])[0]
-        reference_vector = self.model.base.predict(self._create_input_vector(binned_reference))
-        query_vector = self.model.base.predict(self._create_input_vector(binned_query))
-
-        return cosine_similarity(reference_vector[0, :], query_vector[0, :])
+        embedding_reference = self.get_embedding_array([reference])
+        embedding_query = self.get_embedding_array([query])
+        return cosine_similarity(embedding_reference[0, :], embedding_query[0, :])
 
     def matrix(self, references: List[Spectrum], queries: List[Spectrum],
                array_type: str = "numpy",
@@ -122,33 +101,13 @@ def matrix(self, references: List[Spectrum], queries: List[Spectrum],
         ms2ds_similarity
             Array of MS2DeepScore similarity scores.
         """
-        reference_vectors = self.calculate_vectors(references)
+        embeddings_reference = self.get_embedding_array(references)
         if is_symmetric:
             assert np.all(references == queries), \
                 "Expected references to be equal to queries for is_symmetric=True"
-            query_vectors = reference_vectors
+            embeddings_query = embeddings_reference
         else:
-            query_vectors = self.calculate_vectors(queries)
+            embeddings_query = self.get_embedding_array(queries)
 
-        ms2ds_similarity = cosine_similarity_matrix(reference_vectors, query_vectors)
+        ms2ds_similarity = cosine_similarity_matrix(embeddings_reference, embeddings_query)
         return ms2ds_similarity
-
-    def calculate_vectors(self, spectrum_list: List[Spectrum]) -> np.ndarray:
-        """Returns a list of vectors for all spectra
-
-        parameters
-        ----------
-        spectrum_list:
-            List of spectra for which the vector should be calculated
-        """
-        n_rows = len(spectrum_list)
-        reference_vectors = np.empty(
-            (n_rows, self.output_vector_dim), dtype="float")
-        binned_spectrums = self.model.spectrum_binner.transform(spectrum_list, progress_bar=self.progress_bar)
-        for index_reference, reference in enumerate(
-                tqdm(binned_spectrums,
-                     desc='Calculating vectors of reference spectrums',
-                     disable=(not self.progress_bar))):
-            reference_vectors[index_reference, 0:self.output_vector_dim] = \
-                self.model.base.predict(self._create_input_vector(reference), verbose=0)
-        return reference_vectors